python - 生成 Pandas 中排名最高的值的列

Question

我有一个数据框topic_data，其中包含LDA主题模型的输出:

topic_data.head(15)

    topic                      word     score
0       0                Automobile  0.063986
1       0                   Vehicle  0.017457
2       0                Horsepower  0.015675
3       0                    Engine  0.014857
4       0                   Bicycle  0.013919
5       1                     Sport  0.032938
6       1      Association_football  0.025324
7       1                Basketball  0.020949
8       1                  Baseball  0.016935
9       1  National_Football_League  0.016597
10      2                     Japan  0.051454
11      2                      Beer  0.032839
12      2                   Alcohol  0.027909
13      2                     Drink  0.019494
14      2                     Vodka  0.017908

这显示了每个主题的前 5 个术语以及每个术语的分数(权重)。我想要做的是重新格式化，以便索引是术语的排名，列是主题 ID，值是从 word 和 score< 生成的格式化字符串 列(类似于 "%s (%.02f)"% (word,score))。这意味着新的数据框应该如下所示:

Topic  0                1                            ...
Rank
  0  Automobile (0.06)  Sport (0.03)                 ...
  1  Vehicle (0.017)    Association_football (0.03)  ...
 ... ...                ...                          ...

解决这个问题的正确方法是什么？我认为它涉及索引设置、拆栈和排名的组合，但我不确定正确的方法。

Answer 1

它会是这样的，请注意，必须首先生成Rank:

In [140]:
df['Rank']    = (-1*df).groupby('topic').score.transform(np.argsort)
df['New_str'] = df.word + df.score.apply(' ({0:.2f})'.format)
df2           = df.sort(['Rank', 'score'])[['New_str', 'topic','Rank']]
print df2.pivot(index='Rank', values='New_str', columns='topic')

topic                  0                                1               2
Rank                                                                     
0      Automobile (0.06)                     Sport (0.03)    Japan (0.05)
1         Vehicle (0.02)      Association_football (0.03)     Beer (0.03)
2      Horsepower (0.02)                Basketball (0.02)  Alcohol (0.03)
3          Engine (0.01)                  Baseball (0.02)    Drink (0.02)
4         Bicycle (0.01)  National_Football_League (0.02)    Vodka (0.02)